A reference needs to do two jobs: identify a specific work, and let a reader reach it. A DOI and a URL each help with one of those jobs, and the confusion usually starts when a manuscript treats them as interchangeable. They are not. A DOI is a registered identifier that resolves through doi.org to whatever destination the registrant currently maintains. A URL is a locator for a page at a particular address. When you are revising a manuscript, handing a project to a coauthor, or preserving an evidence trail, the right choice depends on which job you need the reference to perform.
The short answer: use the DOI as the stable identifier when the work has one, and use a URL when you are pointing to something that is not identified by a DOI or when you need to record the specific page you consulted. In practice, most reference lists need both kinds of entries, and the discipline is in knowing which is which. This article walks through the decisions, the artifacts to keep, and how to tell when a reference is actually ready.
What problem are you actually solving when you pick a DOI or a URL?
Start by naming the task. If you are building a reference list for a manuscript, the goal is a citation that a reader or reviewer can follow to the same work you used. If you are closing a project, the goal is an evidence trail that a future collaborator can audit. If you are starting a review, the goal is a searchable, reproducible set of sources.
Each of these tasks has a different failure mode. A broken URL fails the reader who wants to reach the page. A DOI that resolves to a landing page with different metadata fails the reader who wants to confirm identity. A reference with neither fails both. So the first decision is not "DOI or URL" but "what does this reference need to prove?"
A useful diagnostic: ask whether the work you are citing has a DOI registered with a metadata agency. Crossref, for example, exposes publisher-deposited scholarly metadata through its REST API, including titles, authors, venues, dates, identifiers, licenses, funding, ORCID and ROR identifiers, abstracts when deposited, and post-publication updates from trusted sources. If the work is in that kind of registry, the DOI is the identifier you want to anchor the reference. If it is not, a URL may be the only locator you have.
How does a DOI work, and what does it actually guarantee?
A DOI resolves through doi.org to the current destination registered for that identifier. That is the key behavior: the identifier stays the same while the destination can change. This is why DOIs are preferred for scholarly works that have them. The publisher can move the landing page, and the DOI still points to the current location.
But resolution is not the same as verification. Resolution helps locate a work, while bibliographic fields and the source text still need to be checked for identity and claim support. A DOI that resolves successfully tells you the identifier is registered and the destination exists. It does not tell you that the title, authors, and venue in your reference match the work, and it does not tell you that the work supports the claim you are making. Those are separate checks.
This distinction matters most during manuscript revision. A coauthor may add a DOI to a reference without checking that the metadata matches. A reviewer may follow the DOI and land on a page that looks right but has a different publication year. The DOI did its job; the reference still needs a metadata check.
When is a URL the right choice?
A URL is the right choice when the thing you are citing is not identified by a DOI, or when you need to record the specific page or version you consulted. Preprints on servers without DOI registration, datasets hosted on a project site, software documentation, and web pages that are not scholarly publications all fall into this category. In those cases, a URL is the locator, and the reference should make clear what the reader will find there.
A URL is also useful as a supplement when you need to record the exact page you used, even if the work has a DOI. The DOI identifies the work; the URL records the access point. Keeping both in your working notes, even if only the DOI appears in the final reference, preserves the trail.
The failure mode for URLs is well known: they break. Pages move, sites restructure, and links rot. That is why a URL-only reference is weaker than a DOI-anchored one when a DOI exists. But when no DOI exists, a URL with a clear description of what it points to is better than nothing.
Where does each option fail on the same research task?
Consider a concrete task: you are revising a manuscript and need to confirm that every reference in your bibliography points to the work you actually cited. Run both options against the same criteria.
| Criterion | DOI | URL |
|---|---|---|
| Stable identifier | Yes, registered and resolvable through doi.org | No, address can change or break |
| Confirms work identity | Only if metadata is checked against the source | Only if the page content is checked |
| Confirms claim support | No, requires reading the source | No, requires reading the source |
| Works for non-scholarly sources | Usually not available | Yes, when a page exists |
| Records exact page consulted | Not by itself | Yes, if the URL is specific |
| Survives publisher restructuring | Yes, destination updates | No, link may break |
The table makes the division of labor clear. The DOI is the identity anchor. The URL is the access point. Neither confirms that the source supports your claim; that requires reading the work. A reference is ready when the identifier resolves, the metadata matches, and you have confirmed the claim against the source text.
What artifacts should you keep for an auditable trail?
An auditable trail is more than a reference list. It is the set of artifacts that let a collaborator reconstruct how you found, selected, and used each source. At minimum, keep:
- The DOI for every work that has one, recorded exactly as registered.
- The URL you actually consulted, especially for non-DOI sources or specific pages.
- The bibliographic fields you verified: title, authors, venue, date.
- A note on which claim in your manuscript each source supports.
- The search or screening steps that led you to the source.
That last item is where many trails break. A reference list shows what you cited, not how you found it. If a coauthor needs to reproduce your search, they need the queries, the databases, and the screening decisions. This is where a structured research workspace helps. You can organize saved papers and their metadata in My Library, and keep the project-level context in Research Projects so the trail lives with the work rather than in a scattered folder.
How do you verify a reference before it goes into a manuscript?
Verification has three layers, and each one catches a different problem.
Layer one: identifier resolution. Follow the DOI through doi.org and confirm it reaches a live destination. For a URL, open it and confirm the page still exists and matches what you cited.
Layer two: metadata match. Compare the title, authors, venue, and date in your reference against the registered metadata. Crossref's API exposes these fields when they are deposited, which makes the comparison concrete rather than impressionistic.
Layer three: claim support. Read the source and confirm it says what you claim it says. No identifier can do this for you.
For a bibliography with many entries, layer two is the bottleneck. This is where citation screening tools earn their place. TrueCite checks BibTeX against real academic databases to flag hallucinated references, which is useful when a coauthor has added entries you have not personally verified. It addresses the metadata layer; the claim-support layer still requires reading.
What does a combined workflow look like from search to final reference?
A combined workflow treats the DOI and the URL as two outputs of the same process, not as competing choices.
- Search and screen. Find candidate sources and record where you found them. Use WisPaper search to locate academic papers, and note the query and date so the search is reproducible.
- Capture identifiers. For each selected source, record the DOI if one exists and the URL you consulted. Do this at selection time, not at manuscript time, when the trail is still fresh.
- Verify metadata. Check title, authors, venue, and date against the registered record. Flag any entry where the metadata does not match.
- Confirm claim support. Read the source and note which claim it supports. This is the step that makes the reference defensible.
- Format the reference. Use the DOI as the identifier when available; use the URL when no DOI exists or when the specific page matters.
- Screen the final bibliography. Run the completed BibTeX through a citation check to catch entries that slipped through.
For longer projects, the search and screening steps can be delegated to a structured agent workflow. Scholar Agent supports longer research tasks in the literature review, inspiration, and reproduction areas, which fits the early stages when you are building the candidate set. Once papers are saved, Library QA lets you ask questions based on papers uploaded to My Library or added from search results, which helps with the claim-support layer when you need to check what a source actually says.
How do you know the reference is ready?
A reference is ready when three conditions hold: the identifier resolves, the metadata matches the registered record, and you have confirmed the claim against the source text. If any of these is missing, the reference is a draft.
A practical readiness check for a manuscript:
- Every DOI resolves through doi.org.
- Every DOI's metadata matches the reference entry.
- Every URL still opens and matches the cited page.
- Every source has a note on which claim it supports.
- The search and screening steps are recorded somewhere a coauthor can find them.
If you can answer yes to all five, the reference list is auditable. If not, you know exactly which layer needs work.




